Parv Kapoor
@Parvkpr
Followers
186
Following
256
Media
23
Statuses
69
PhD-ing at CMU + interning at @msftresearch || previously @genrobotics_ai, USC, and @CompScienceCU
Pittsburgh, PA
Joined February 2023
Woke up to some good news that STLCG++ was accepted to IEEE Robotics and Automation Letters (RA-L), 2025. If you’re interested in neurosymbolic methods, please check out our library! .Super thankful to my collaborators Karen Leung, Kazuki Mizuta and @eskang09 for their support!.
Thrilled to share our work on accelerating neurosymbolic methods for robotics!. STLCG++ encodes logical constraints in a computational graph, achieving 100-1000× faster quantitative evaluation—enabling real-time symbolic reasoning for traj opt, policy learning, & more. 🧵 [1/n]
0
0
6
I saw it respond to commands in hindi (pick up the grapes) and was very pleasantly surprised
Come by the @GoogleDeepMind booth at @RoboticsSciSys conference in LA!. We’re demoing Gemini Robotics On-Device live, come check it out
0
0
4
come checkout our demo on how our setup does precise detections for far away agents🕵️ that are sometimes invisible to the naked eye
Catch our team @Parvkpr @PatrikarJay @AirLabCMU presenting and demoing ViSafe at #RSS2025 tomorrow!. We'll be showing our payload demo & high speed aerial collision avoidance results 🚀.
0
1
6
RT @Nik__V__: Catch our team @Parvkpr @PatrikarJay @AirLabCMU presenting and demoing ViSafe at #RSS2025 tomorrow!. We'll be showing our pay….
0
3
0
if you’re attending RSS 2025 in LA, Let’s catch up! I’d love to hear what everyone’s been working on. (P.S.: my talk is tomorrow, please come attend and hear about our experience building and field testing full stack solutions for drone autonomy) 🚀
0
0
3
RT @akapoor_av8r: Scaled Foundations is now General Robotics. We’re building general-purpose intelligence for every robot. Across any scen….
0
16
0
I’ll be interning at Microsoft Research AI Frontiers team this summer in the big 🍎. I’ll be working on large action models and reinforcement learning for safe AI! If you’re in nyc, hmu!.
0
0
10
Super proud to share our new RSS 2025 work! I have been working on this with my collaborators for a while now and the sheer complexity of deploying CBFs in the field with perceptual inputs has been so exciting! Come chat with us about it at RSS!.
🚀 Thrilled to present ViSafe, a vision-only airborne collision avoidance system that achieved drone-to-drone avoidance at 144 km/h. In an era of congested airspace and growing autonomy, reliable self-separation is paramount 🧵👇
0
1
8
RT @YiningShe: 🚀 Excited to present our work at #ICSE2025 next week in Ottawa!. FairSense: Long-term Fairness Analysis of ML-enabled System….
0
2
0
Our paper introduces ATLAS, a new framework for learning specifications from examples with.user-specified structural constraints -enabling much more controlled and practical specification.mining, repair, synthesis. If you’re attending, let’s grab a coffee and chat! ☕
0
0
1
(fun fact: This year ICSE has 4000 authors, 483.sessions, 1600 presentations!) Please come attend my talk!:.(1) Constrained LTL Specification Learning from Examples.🗓 Wednesday (April 30th), 16:15–16:30, 📍Room 103. 📄 Preprint: 2/N 🪡.
1
0
1
Open-Source & Ready to Use! STLCG++ is available in JAX & PyTorch for easy integration into ML & robotics workflows. Try our STLCG++ trajectory optimization demo in GRID! 🔗. #Robotics #AI #DifferentiableProgramming #Neurosymbolic.[8/n]
1
0
0
Impact for robotics?.Before STLCG++, STLCG was too slow for large-scale AI & robotics. Now, STLCG++ enables:.🚗Autonomous driving safety (real-time monitoring) 🦾Robot learning (STL as a structured loss function).📊Data generation (better training data for VLAs). [7/n].
1
0
0
Real-World Robotics Applications 🌍. We applied STLCG++ for:. ✅ Trajectory Optimization: Finds optimal paths satisfying STL constraints. ✅ STL-Guided Diffusion: achieving 62.5% success for policies vs. unguided models. ✅ STL Spec Mining: Recovers STL specs from data. [6/n]
1
0
1
Major advantages of STLCG++:.✅ 100-1000× faster robustness evaluation on GPU, enabling online STL .✅ Efficient differentiation for gradient-based optimization over time.✅ GPU acceleration with JAX & PyTorch, scaling to long-horizon tasks.STL 🤝 complex robotic behavior!. [5/n]
1
0
0